Scholarlink

<p>Alzheimer’s disease is one of the most common degenerative brain diseases, and its early detection plays an important role in improving clinical management, increasing treatment effectiveness, and reducing the consequences of disease progression. In this study, a family of convolutional neural network architectures, CortexNet, was designed and evaluated to investigate the effect of increasing the depth of the architecture on detection performance and computational complexity. After comparing different versions, the CortexNet_V1 architecture was selected as the final model. To evaluate the reliability and stability of the model, the training and testing process was performed in three independent runs with random seed values of 42, 123, and 456, and the model performance was examined in three two-class, three-class, and four-class classification scenarios. The average model accuracy in these three scenarios was 98.91 ± 0.31%, 98.60 ± 0.21%, and 98.39 ± 0.36%, respectively, indicating the good stability of the model against different initialization. In addition, the proposed architecture, with only 0.059 GFLOPs, has much lower computational complexity compared to deeper CortexNet versions and pre-trained models, while maintaining its competitive performance. The results of this study show that for the used dataset, a lightweight and targeted architecture can significantly reduce the computational cost without the need to increase the network depth, while maintaining high accuracy, and provide a suitable option for the development of intelligent systems for the diagnosis of Alzheimer’s disease, especially in environments with limited computational resources.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

CortexNet: convolutional neural networks for alzheimer’s disease diagnosis using brain MRI

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月31日

查看全文
  • Hosna Ghahramani,
  • Mahdiyeh Velaei,
  • Fardin Piroozi,
  • Ali Ghaffari,
  • Laya Yari

摘要

Alzheimer’s disease is one of the most common degenerative brain diseases, and its early detection plays an important role in improving clinical management, increasing treatment effectiveness, and reducing the consequences of disease progression. In this study, a family of convolutional neural network architectures, CortexNet, was designed and evaluated to investigate the effect of increasing the depth of the architecture on detection performance and computational complexity. After comparing different versions, the CortexNet_V1 architecture was selected as the final model. To evaluate the reliability and stability of the model, the training and testing process was performed in three independent runs with random seed values of 42, 123, and 456, and the model performance was examined in three two-class, three-class, and four-class classification scenarios. The average model accuracy in these three scenarios was 98.91 ± 0.31%, 98.60 ± 0.21%, and 98.39 ± 0.36%, respectively, indicating the good stability of the model against different initialization. In addition, the proposed architecture, with only 0.059 GFLOPs, has much lower computational complexity compared to deeper CortexNet versions and pre-trained models, while maintaining its competitive performance. The results of this study show that for the used dataset, a lightweight and targeted architecture can significantly reduce the computational cost without the need to increase the network depth, while maintaining high accuracy, and provide a suitable option for the development of intelligent systems for the diagnosis of Alzheimer’s disease, especially in environments with limited computational resources.